Biomedical Engineering Reference
In-Depth Information
Fig. 3.
Histograms of
−log
10
(p-value) values resulting from association tests of the phenotype
with the causal SNP ancestor nodes (As) and the other latent nodes (Os). General results compil-
ing all simulated scenarii (see 6.1): MAF (0.1-0.2, 0.2-0.3 and 0.3-0.4), heterozygous GRR (
1
.
4
,
1
.
6
and
1
.
8
) and disease model (dominant, recessive, additive and multiplicative)
from the association test between variable
X
v
(belonging to layer
)and
Y
. Thus can
be assessed association significance, corrected for family-wise type I error, or should
one write instead, controlled for
layer-wise
type I error.
7
Detection of Indirect Genetic Associations
7.1
Verification for Simulated Data
Over all
3600
runs (
36
scenarii
100
replicates), up to
7
layers were generated. Results
obtained for layers above third layer (comprised) are not reported: such layers do not
provide sufficient data to compute representative medians or draw informative boxplots.
On average, over all
3600
FLTMs, the percentages of nodes are distributed as follows:
89
.
1%
in layer
0
,
9
.
5%
in layer
1
,
1
.
2%
in layer
2
and
0
.
2%
in layer
3
.
×
log
10
(p-value) values. Val-
ues near
0
point out independence and the previous indicator increases with the strength
of the dependence. Figure 3 compares the histograms of
−
General Trends.
Instead of p-values, we will consider the
−log
10
(p-value) values result-
ing from association tests of
Y
with A nodes and O nodes, respectively. The comparison
of these two histograms reveals a large dissimilarity between the two distributions. The
majority (
70%
)of
log
10
(p-value) values relative to A nodes is greater than
1
, whereas
it is the case for only
19%
of O nodes. We observe that large
−
log
10
(p-value) val-
ues (
e.g.
, greater than
5
) are common for the former and are very rare for the latter. A
non-parametric test, the Wilcoxon rank-sum test, shows a p-value less than
10
−
16
, thus
confirming that A and O p-values follow two different distributions.
Figure 4(a) more thoroughly describes the
−
log
10
(p-value) values observed for the
different layers of the FLTM in the cases of tests relative to A and O nodes. The layer
0
refers to the association tests between the phenotype and the causal SNP and serves
as the reference value. In this figure, we observe that the association strength for A
nodes slowly decreases when the layer number increases, whereas the association for
O nodes sticks to
−
log
10
(p-value) values below
0
.
4
, corresponding to p-values greater
than
0
.
4
. Although O nodes reveal false positive associations (less than
10%
have a p-
value below
0
.
01
), these results clearly highlight a general trend: indirect associations
are captured by the A nodes while it is not the case for a large majority of O nodes.
−
Search WWH ::
Custom Search